Further Improvements to the Statistical Hurricane ...rammb.cira.colostate.edu/resources/docs/further_demaria.pdfA verification of the SHIPS operational intensity forecasts is presented

Further Improvements to the Statistical Hurricane Intensity PredictionScheme (SHIPS)

MARK DEMARIA

NOAA/NESDIS, Fort Collins, Colorado

MICHELLE MAINELLI

Tropical Prediction Center, Miami, Florida

LYNN K. SHAY

RSMAS/MPO, University of Miami, Miami, Florida

JOHN A. KNAFF

Colorado State University/CIRA, Fort Collins, Colorado

JOHN KAPLAN

NOAA/Hurricane Research Division, Miami, Florida

(Manuscript received 21 April 2004, in final form 27 October 2004)

ABSTRACT

Modifications to the Atlantic and east Pacific versions of the operational Statistical Hurricane IntensityPrediction Scheme (SHIPS) for each year from 1997 to 2003 are described. Major changes include theaddition of a method to account for the storm decay over land in 2000, the extension of the forecasts from3 to 5 days in 2001, and the use of an operational global model for the evaluation of the atmosphericpredictors instead of a simple dry-adiabatic model beginning in 2001.

A verification of the SHIPS operational intensity forecasts is presented. Results show that the 1997–2003SHIPS forecasts had statistically significant skill (relative to climatology and persistence) out to 72 h in theAtlantic, and at 48 and 72 h in the east Pacific. The inclusion of the land effects reduced the intensity errorsby up to 15% in the Atlantic, and up to 3% in the east Pacific, primarily for the shorter-range forecasts. Theinclusion of land effects did not significantly degrade the forecasts at any time period. Results also showedthat the 4–5-day forecasts that began in 2001 did not have skill in the Atlantic, but had some skill in the eastPacific.

An experimental version of SHIPS that included satellite observations was tested during the 2002 and2003 seasons. New predictors included brightness temperature information from Geostationary OperationalEnvironmental Satellite (GOES) channel 4 (10.7 �m) imagery, and oceanic heat content (OHC) estimatesinferred from satellite altimetry observations. The OHC estimates were only available for the Atlanticbasin. The GOES data significantly improved the east Pacific forecasts by up to 7% at 12–72 h. Thecombination of GOES and satellite altimetry improved the Atlantic forecasts by up to 3.5% through 72 hfor those storms west of 50°W.

Corresponding author address: Mark DeMaria, NOAA/NESDIS/ORA, CIRA/CSU, W. Laporte Ave., Fort Collins, CO 80523.E-mail: [email protected]

AUGUST 2005 D E M A R I A E T A L . 531

© 2005 American Meteorological Society

WAF862

1. Introduction

The National Hurricane Center (NHC) has a suite ofskillful global and regional prediction models availableas guidance for tropical cyclone track prediction. How-ever, operational intensity models have considerablyless skill than the track models (DeMaria and Gross2003). For intensity, the three primary models arethe simple Statistical Hurricane Intensity Forecast(SHIFOR) model (Jarvinen and Neumann 1979; Knaffet al. 2003), which uses climatology and persistence tomake a prediction; the Statistical Hurricane IntensityPrediction Scheme (SHIPS), which includes storm en-vironmental conditions in addition to climatology andpersistence; and the National Centers for Environmen-tal Prediction (NCEP) version of the Geophysical FluidDynamics Laboratory (GFDL) hurricane model (Kuri-hara et al. 1998). The SHIFOR model is primarily usedas a benchmark for evaluating the skill of the officialNHC intensity forecasts and those from the objectivemodels. The GFDL model became operational in 1995,and a version that uses input from the U.S. Navy globalmodel became available at NHC in 1999. The purposeof this paper is to document recent changes to theSHIPS model, evaluate its performance relative to theSHIFOR and GFDL models, and describe a recent ef-fort to reduce the skill gap between the SHIPS intensitymodel and track models through the inclusion of newpredictors from satellite observations.

The SHIPS model has been run operationally forstorms in the Atlantic basin since 1991 and in the eastPacific basin since 1996. The model characteristics andits performance through 1997 were documented in De-Maria and Kaplan (1994a) and DeMaria and Kaplan(1999, hereafter DK99). Since 1997 significant modifi-cations have been made including the addition of anadjustment factor to account for the decay over land,the extension of the forecasts from 3 to 5 days, andchanges to the large-scale model from which many ofthe predictors are derived. Over the past several yearsSHIPS has become one of the primary guidance modelsthat NHC uses for their operational intensity forecasts.For this reason there is a need to document the modelchanges relative to the version described by DK99 andto perform a verification of the operational SHIPS in-tensity forecasts.

The predictors for SHIPS include climatology andpersistence, atmospheric environmental parameters(vertical shear, etc.), and sea surface temperature(SST), but contain relatively little information aboutthe storm itself. Fitzpatrick (1997) showed that statisti-cal intensity forecasts for western North Pacific tropicalcyclones could be improved by including predictors

from infrared satellite data. In this paper, the potentialfor improving SHIPS using predictors from Geostation-ary Operational Environmental Satellite (GOES) infra-red (10.7 �m) imagery will be evaluated

The ocean influence on intensity change is includedin SHIPS by evaluating the SST at the storm center.The SST is determined from the weekly 1° latitude–longitude analyses described by Reynolds and Smith(1993). The SST is then used to estimate the maximumpotential intensity (MPI) from the empirical formuladeveloped by DeMaria and Kaplan (1994b) for the At-lantic and by Whitney and Hobgood (1997) for the eastPacific. Many studies have shown that the upper-oceanheat content through the depth of the 26°C isotherm isimportant for tropical cyclone intensity change (e.g.,Shay et al. 2000), and that the SST is not always a goodmeasure of the total heat content. A method to esti-mate the oceanic heat content (OHC) from EuropeanRemote Sensing Satellite-2 (ERS-2) and TOPEX/Poseidon satellite altimetry observations was devel-oped by Mainelli-Huber (2000), and an archive of oce-anic heat content analyses for the prestorm environ-ment of most Atlantic tropical cyclones from 1995 tothe present was produced. Real-time OHC analyseswere implemented at NHC beginning in 2002. The pos-sibility of improving the SHIPS forecasts using OHCpredictors will also be evaluated.

The modifications to the SHIPS model from 1997 to2003 are described in section 2, and an evaluation of themodel performance is presented in section 3. In section4, the evaluation of a parallel version of SHIPS withpredictors from GOES and satellite altimetry data isdescribed.

2. The operational SHIPS model

The 1997 operational version of the SHIPS modelwas described in DK99. Beginning that year, SHIPSwas converted to a statistical–dynamical model wheresome of the atmospheric predictors were determinedfrom a numerical forecast model. A multiple linear re-gression was used to develop the model equations,where the dependent variable is the intensity changefrom 0 to 12, 0 to 24, . . . , or 0 to 72 h (separate regres-sions for each time interval), and the independent vari-ables are parameters believed to be important for in-tensity change. All predictors must be significant at the1% level (using a standard F statistic) for at least oneforecast period. For forecast consistency, the same setof predictors is used at all forecast intervals. The inclu-sion of predictors that are not significant at the 1%level does not present a problem because all of theincluded predictors are significant at more than half of

532 W E A T H E R A N D F O R E C A S T I N G VOLUME 20

the forecast intervals (e.g., for the 2003 model 83% ofthe coefficients passed the 1% significance test), andthe magnitudes of the predictors tend to become smallat the forecast intervals when they are not significant.The model for the east Pacific uses the same predictorsas for the Atlantic version but with different coeffi-cients. The model has undergone significant changessince 1997 as summarized in the remainder of this sec-tion.

a. Dependent sample

The 1997 model was developed from all storm casesfrom 1989 to 1996 that reached at least tropical stormstrength. The sample was restricted to cases where thestorm track remained over water for the entire intervalfor which the dependent variable was calculated. Foreach year since 1997 the cases from the previous yearwere added to the sample and the prediction coeffi-cients were rederived. The sample for the 2003 seasonincluded all Atlantic storms from 1989 to 2002. As willbe described in section 3, the rules that NHC uses toperform their annual verifications have changed overthe years. The dependent sample for SHIPS was modi-fied to remain consistent with these verification rules.For this purpose tropical depressions that never gotstrong enough to be named were added in 2001 andsubtropical storms were added in 2003.

b. Track changes

The real-time forecasts in 1997 used the track fromthe operational Limited Area Sine Transform Barotro-pic (LBAR) model (Horsfall et al. 1997). This choicesometimes presented a problem because the LBARtrack was not always consistent with the NHC officialforecast track. Beginning in 1999 the LBAR track wasreplaced by the official NHC track forecast. Becausethe SHIPS model is usually run before the NHC trackforecast is completed, the NHC track forecast from theprevious forecast cycle (6 h old, but corrected to matchthe current storm position) is used. Beginning in 2001the official NHC forecasts were extended from 72 to120 h on an experimental basis. In 2003 the 5-day fore-casts became operational. SHIPS was modified to pro-vide intensity forecasts out to 120 h beginning in 2001.

c. Land correction

The SHIPS developmental sample excludes casesthat crossed major landmasses. Beginning in 2000 thereal-time forecasts over land were modified in a post-processing step using the simple empirical decay modeldescribed by Kaplan and DeMaria (1995). The land

modification procedure begins with the SHIPS predic-tion without land effects. The forecast positions andintensities (every 12 h) are interpolated to 1-h intervals,and the overland portions of the track are identified.The empirical decay model is then applied to the part ofthe track that is over land. If the forecasted stormmoves back over the water, the remaining portion ofthe prediction is adjusted so that the intensity changesare the same as for the unadjusted forecast. It should bepointed out that this method introduces some error forcases that move back over the water because the re-mainder of the prediction following the landfall is basedupon statistical relationships that did not take into ac-count the reduction in the intensity due to interactionwith land. A method is being developed to correct thisproblem by reformulating the model with dependentvariables that are the intensity change in 12-h intervals(0–12, 12–24, etc.) rather than from zero to the forecasttime (0–12, 0–24, etc.).

Figure 1 shows the original and adjusted SHIPS 72-hforecasts for Hurricane Isidore beginning at 1200 UTCon 18 September 2002. The forecast track began in thewestern Caribbean, moved over western Cuba from 48to 55 h, and then moved into the Gulf of Mexico. Thedecay model reduces the maximum winds by a factor of0.9 at landfall to account for land � sea differences insurface roughness, and then applies an exponential de-cay formula for the portion of the track over land.When the track point is south of 36°N, the decay coef-ficients for the southeastern United States are appliedand for points north of 40°N the coefficients for theNew England area (Kaplan and DeMaria 2001) areused. A linear combination of the coefficients is used

FIG. 1. The original SHIPS forecast and the forecast adjustedfor movement over land for Hurricane Isidore beginning at 1200UTC on 18 Sep 2002. Isidore crossed western Cuba between 48and 55 h.


between these two latitudes. When the storm movesback over the water, the intensity is divided by 0.9,which accounts for the intensity increase at 55 h in theadjusted forecast in Fig. 1. Because the real-time trackforecasts are not perfect, both the original (unadjusted)and adjusted SHIPS intensity predictions are providedto the forecasters. The forecasts adjusted for decay overland will be referred to as D-SHIPS.

d. Model fields

The SHIPS model utilizes a “perfect prog” approach(Kalnay 2003) where the predictors are calculated fromanalyses along the best-track positions in the dependentsample to develop the coefficients but applied using theNCEP global model along a forecast track in real time.For the 1997 model, operational analyses were used forthe model development. Beginning in 2003 the opera-tional analyses from 1989 to 2000 were replaced by theNCEP–National Center for Atmospheric Research(NCAR) reanalyses to obtain a more complete andconsistent dataset.

When SHIPS was converted to a statistical–dynamical model in 1997, atmospheric predictors forthe real-time runs were determined from a simple 10-layer dry-adiabatic prediction model where the stormcirculation was filtered from the initial condition. Thisfiltering procedure eliminated the difficulty of separat-

ing the storm from its environment during the modelforecast. Because the 10-layer model had very limitedaccuracy, the predictors from this model were used onlyout to 48 h. When the SHIPS forecasts were extendedto 5 days in 2001, the adiabatic model fields out to 48 hwere replaced by 5-day forecasts from the NCEP globalforecasting system (GFS).

e. Predictor changes

The SHIPS forecasts are verified at the end of everyhurricane season. Based upon the model performance,modifications are made to the predictors almost everyyear. New predictors are tested if there is some physicalreasoning behind them and they are added only if theypass the underlying 1% statistical significance test fortheir coefficients. Predictors are sometimes removedeach year because they no longer pass the significancetest. Modifications to the SHIPS predictors since 1997are summarized below, and the physical reasoning forthe new predictors is briefly described.

Some of the SHIPS predictors such as the initialstorm intensity are “static,” in that they are evaluatedonly at t � 0 h. The same value of a static predictor isused as input to the intensity change forecast for eachtime interval. Other predictors such as the verticalshear are time dependent, where they are averagedalong the storm track. Table 1 summarizes the predic-

TABLE 1. Predictors used in SHIPS 1997–2003. The predictors that are evaluated at the beginning of the forecast period are static (S),and predictors that are evaluated along the forecasted track of the storm are time dependent (T). An X indicates that the predictor wasused in that year and a dash (—) indicates it was not used that year.

PredictorStatic (S) or time

dependent (T)

Year

1997 1998 1999 2000 2001 2002 2003

1) Absolute value of (Julian day � peak season value) S X X X X X X —2) Gaussian function of (Julian day � peak value) S — — — — — — X3) Initial maximum winds S — X X X X X X4) Max wind change during the past 12 h S X X X X X X X5) Initial max winds times previous 12-h change S — — — — — — X6) Zonal component of storm motion S — X X X X X X7) Pressure level of storm steering S — — — — X X X8) 200-hPa divergence S — X X X — X X9) 200-hPa eddy momentum flux convergence S — — — — — X —

10) 200-hPa eddy momentum flux convergence T X X X X X — —11) Max potential intensity � current intensity T X X X X X X X12) 850–200-hPa vertical shear T X X X X X X X13) 200-hPa zonal wind T X X X X X X —14) 200-hPa temperature T X X X X X X X15) 850–700-hPa relative humidity T — — — — X X —16) 500–300-hPa relative humidity T — — — — — — X17) 850-hPa relative vorticity T X X X X X X X18) Surface � 200-hPa �e deviation of lifted parcel T — — — — — — X19) Vertical shear times sine of storm latitude T X X X X X X X20) Square of potential � current intensity T X X X X X X X21) Initial intensity time shear T — — — — — X X


tors used in the operational SHIPS model from 1997 to2003.

The 1997 predictors are described in detail in DK99.In 1998, three new static predictors were added: theinitial maximum wind, the zonal component of thestorm motion, and the 200-hPa divergence. The initialmaximum wind provides information about the organi-zational stage of the storm. The zonal component ofmotion distinguishes between storms in easterly versuswesterly basic currents. The divergence was added as ameasure of synoptic-scale forcing. The divergencecould have been included as a time-dependent predic-tor, but tests showed that there was a large differencebetween the perfect prog divergence and the diver-gence from the model forecasts. The 1998 predictorswere also used in 1999 and 2000.

In 2001, one static predictor (the 200-hPa diver-gence) was eliminated and one was added, and onetime-dependent predictor was added. During the 2000Atlantic season there were two storms (Debby andJoyce) that experienced strong vertical shear and ap-peared to decouple in the vertical based upon theirappearance in GOES satellite imagery. Although theenvironmental conditions later became more favorable,these two storms never reintensified. It also appearedthat these storms began to move with the low-level flowafter they decoupled in the vertical. To help account forhighly sheared systems, a method was developed to findthe weights on the horizontal winds from 1000 to 150hPa surrounding the storm (200–800-km annular aver-age) that provide the best match to the observed stormmotion. The center of mass determined by theseweights is referred to as the steering layer pressure(SLP). For storms that became decoupled in the verti-cal, the SLP occurs at a higher pressure, and it wasfound that this variable is negatively correlated withintensity change. Dunion and Velden (2004) used mul-tispectral imagery from GOES to identify the Saharaair layer (SAL), which is characterized by very dry airin the 850–500-hPa layer. Their results show that theSHIPS model overforecasted intensification for stormsthat interacted with the SAL (including Joyce andDebby). To represent SAL effects in SHIPS, a low-level (850–700 hPa) relative humidity predictor fromthe NCEP global model was added. Although the re-sults of Dunion and Velden suggest that the NCEPglobal analysis does not always represent the magni-tude of the low-level drying associated with the SAL,this predictor was found to be statistically significant.An upper-level (500–300 hPa) humidity predictor wasalso tested but was not significant.

Only minor modifications were made to the predic-tors in 2002. The 200-hPa momentum flux was con-

verted from a time-dependent to a static predictor be-cause the real-time estimates of the parameter weremuch larger than those from the perfect prog variables.This inconsistency may have been due to changes in theresolution of the GFS that were implemented that year.This change to a static predictor was implemented inmid-August of 2002 when the problem was first identi-fied. The divergence predictor was again included in2002 since it passed the significance test. One new pre-dictor (the product of the initial intensity and shear)was added because the response of a strong storm toshear is different than that of a weak storm.

Several changes were made to the predictors in 2003.The 200-hPa momentum flux and zonal wind compo-nent were eliminated, one static predictor was addedand one was modified, and one time time-dependentpredictor was modified and another was added. Thenew static predictor was the product of the previous12-h intensity change and the current intensity, whichhelps to reduce the contribution of the persistence termfor storms that are already very intense. The form ofthe static Julian day predictor was modified because theoriginal form overly penalized very early and very lateseason storms. The new Julian day predictor is given byexp{-[(Jd-Pd)/Rd]2}, where Jd is the Julian day, Pd is thepeak day of the hurricane season from climatology (day253 for the Atlantic, day 238 for the east Pacific), andRd is a time scale that controls the width of the influ-ence of this term. By experimentation it was found thatRd � 25 days provided the best fit for the Atlantic andeast Pacific cases. The time-dependent low-level hu-midity parameter was replaced by the upper-level hu-midity variable. This change was related to the use ofthe reanalysis fields in the dependent sample, whichhave different moisture properties than some of theolder operational NCEP analyses. The poor perfor-mance of SHIPS for Tropical Storm Dolly in 2002 (andsimilar storms in the sample), which did not intensifydespite what appeared to be a low-shear environmentwith fairly warm water, motivated the testing of a newtime-dependent thermodynamic predictor. For thispurpose the surface equivalent potential temperature(�e)is calculated and then assumed to be constant for alifted parcel. The positive differences between the �e ofthis lifted parcel and that of the original NCEP profileare then averaged from 1000 to 200 hPa. This variableis a crude representation of the atmospheric stability.

f. Averaging areas

The atmospheric predictors in Table 1 are averagedover horizontal areas. In the 1997 model the verticalshear and momentum fluxes were averaged from r � 0to 600 km, and the 200-hPa temperature, zonal wind,


and divergence, and the 850-hPa vorticity were aver-aged from r � 0 to 1000 km. These variables are meantto measure the storm environment rather than thestorm itself. This separation was not a problem whenthe 10-layer model was used for these predictors be-cause the storm circulation was filtered from the initialand forecast fields. However, when the 10-layer modelwas replaced by the GFS in 2001, the separation be-tween the storm circulation and the environment be-came problematic because the global model contains arepresentation of most tropical cyclones. This problemis further complicated by the fact that SHIPS uses theNHC official forecast track, which can differ signifi-cantly from the storm track in the GFS. The verticalshear is especially sensitive to these differences becausethe storm will make a substantial contribution to theshear if the center point for the calculation is not lo-cated at the storm center. To help alleviate this prob-lem, the area used to calculate all of the atmosphericpredictors (except 200-hPa divergence and momentumfluxes and 850-hPa vorticity) was modified to an annu-lus from a radius of 200–800 km. The size of the innercircle was based upon the typical differences betweenthe official and GFS tracks out to 3 days. The radii fordivergence and vorticity were maintained at 0–1000 kmbecause these are even more sensitive to the storm cir-culation. The radii for momentum flux was kept at0–600 km because the dependent sample showed thatthere is no relationship between the fluxes outside of600 km and intensity change.

3. Validation of operational forecasts

A database of real-time forecasts from all guidancemodels (track and intensity) is maintained by NHC. A“best track” of positions and intensities is also pro-duced from a postanalysis of all available information.Unless otherwise indicated, all verifications describedbelow used the NHC model forecast and best-track ar-chives. The intensity forecasts are evaluated by com-puting the mean absolute error in knots. The SHIPSforecast errors will also be compared to those from theSHIFOR and GFDL models, and to the official NHCintensity forecasts. The significance of the differencebetween average errors from different models is deter-mined by the method described in DK99. Standard sta-tistical procedures for comparing the means of two datasamples are applied, where the sample size is reducedto account for serial correlation. The 95% level is thethreshold for statistical significance.

SHIFOR uses climatology and persistence to forecastintensity, and will be used as a benchmark for evaluat-ing the skill of other forecasts. If the average errors

from a given model are smaller than those fromSHIFOR, then that model is considered to be skillful.The relative error (RE) defined by

RE � 100�Eshifor � Emodel��Eshifor, �1�

where Emodel and ESHIFOR are the mean absolute errorfrom a given model and SHIFOR, respectively, will beused as a quantitative measure of skill. Here, RE is thepercentage reduction of the model errors relative toSHIFOR, where positive values indicate forecast skill.When the NHC forecasts were extended to 5 days on atest basis in 2001, SHIFOR was updated and extendedfrom 3 to 5 days (Knaff et al. 2003). The updatedSHIFOR model will be referred to as SHIFOR5.

NHC performs a yearly validation of their officialforecasts and their operational guidance models. Formany years NHC restricted their verification sample tocases of at least tropical storm strength (34 kt), andextratropical and subtropical cases were excluded. Be-ginning in 2002, NHC expanded their verification rulesto include all systems that were classified as at least atropical depression in the best track. In addition, NHCbegan naming subtropical storms in 2002, using thesame list of names previously restricted to purely tropi-cal systems. For this reason, subtropical storms are in-cluded in the verification, but extratropical systems arestill excluded. In the discussion below, the traditionalNHC sample selection procedure will be referred to asthe “old rules,” and the procedure implemented in 2002(with depression stage and subtropical systems added)will be referred to as the “new rules.” The official NHCforecast is available at 12, 24, 36, 48, 72, 96, and 120 h,and all verification results are restricted to these timeseven though the SHIPS forecasts are also available at60, 84, and 108 h.

To evaluate the long-term trends in the SHIPS skill,the Atlantic forecasts were verified using the old NHCrules for the periods 1991–96 and 1997–2003. These pe-riods were chosen because SHIPS was converted to astatistical–dynamical model in 1997. The official NHCand the SHIFOR intensity forecasts were also includedfor comparison and for the calculation of RE.

Figure 2 shows that SHIPS did not have any signifi-cant skill at any forecast interval for the 1991–96sample. In contrast, SHIPS has significant skill at alltime intervals out to 72 h for the 1997–2003 sample.This result indicates that the SHIPS modifications de-scribed above have improved the performance of themodel. Figure 2 also shows that the NHC official inten-sity forecasts for the 1991–96 sample were skillful onlyout to 36 h, but were skillful out to 72 h for the 1997–2003 sample. Thus, over the last several years, NHC hasdemonstrated improved intensity forecast skill, which is


probably due to the availability of skillful model guid-ance.

Figure 2 also shows the skill of NHC official trackforecasts [RE as determined by comparison of the av-erage track errors to those from the climatology andpersistence (CLIPER; Neumann 1972) track model].For the 1997–2003 sample, the 12-h NHC track andintensity skill are comparable. By 36 h, the track skill istwice that of the intensity skill, and by 72 h it is threetimes greater. Thus, although some progress has beenmade in the ability to forecast tropical cyclone intensitychange over the last decade, there is still a long wayto go.

Figure 3 shows the RE of the SHIPS and NHC fore-casts for the east Pacific with the old verification rules.It was not possible to show the RE for 1991–96 becauseSHIPS for the east Pacific was not developed until 1996.The SHIPS skill for the east Pacific is generally less

than that for the Atlantic, and was statistically signifi-cant only at 48 and 72 h. The skill of the NHC officialtrack and intensity forecasts is also less than that for theAtlantic. These reductions in skill are probably due tothe fact that the track and intensity of the east Pacificstorms are generally better behaved than in the Atlan-tic and, thus, are easier to predict using climatology andpersistence. Figure 3 also shows that, similar to the At-lantic, the track forecast skill is much greater than theintensity skill except at 12 h.

Figure 4 shows the yearly average SHIPS and NHCofficial (OFCL) intensity forecast errors from 1991 to2003. As shown in Fig. 2, the skill of the SHIPS andNHC Atlantic intensity forecasts improved from 1991to 2003. Figure 4 shows that there is also a downwardtrend in the mean absolute error, although it is notquite as obvious as the skill increase in Fig. 2. As de-scribed by McAdie and Lawrence (2000), it is some-times difficult to detect long-term trends in tropical cy-clone forecasts because the forecast difficulty variesfrom year to year. One way to account for forecastdifficulty is to normalize the errors with respect to cli-matology and persistence forecasts. The relative errordefined by (1) includes this normalization. However,even without the normalization, the slopes of the lineartrend lines in Fig. 4 for SHIPS (OFCL) are �0.39,�0.42, and �0.39 (�0.33, �0.32, and �0.47) kt yr�1 at24, 48, and 72 h, indicating that the intensity forecastshave improved with time.

The only other intensity guidance model routinelyavailable for the Atlantic and east Pacific is the GFDLprimitive equation hurricane model, which became op-erational in 1995 (Kurihara et al. 1998). Figure 5 showsthe relative error of the SHIPS and GFDL intensityforecasts for a homogeneous sample of cases from 1997to 2003, using the old verification rules. The GFDL

FIG. 2. The skill (relative to SHIFOR) of the SHIPS and officialNHC intensity forecasts (OFCL-I) for 1991–96 and 1997–2003Atlantic samples. The skill (relative to CLIPER) of the officialNHC track forecasts (OFCL-T) is shown for reference. Solid(open) symbols indicate that the skill was (was not) statisticallysignificant at the 95% level.

FIG. 3. Same as in Fig. 2 but for the eastern North Pacific forthe 1997–2003 sample.

FIG. 4. The yearly average 24-, 48-, and 72-h intensity errors(kt) for the official NHC and SHIPS forecasts, 1991–2003.


model did not have any forecast skill in either the eastPacific basin or the Atlantic basin through 48 h. TheGFDL model had some skill at 72 h in the Atlantic,although it was not statistically significant at the 95%level. The lack of intensity skill of the GFDL modelindicates the difficulty of initializing a primitive equa-tion model with a strong circulation and where bound-ary layer and convective processes are of first-orderimportance. A number of GFDL model improvementsare being implemented including new parameterizationschemes and increased horizontal and vertical resolu-tion. Thus, the lack of skill in Fig. 5 might not be rep-resentative of future versions of the GFDL model.

A method to account for the storm decay after land-fall was added to SHIPS in 2000 (D-SHIPS). Althoughit might seem obvious that including the decay overland would improve the intensity forecasts, the opera-tional forecast tracks are not perfect. Thus, there areerrors in the timing and location of landfall. Also, theempirical inland decay model is a highly simplified rep-resentation of the physical processes that occur. Figure6 shows the improvement of the D-SHIPS forecastsrelative to the SHIPS forecasts for the Atlantic and eastPacific for 2000–03. For this comparison the new NHCverification rules were applied to increase the numberof cases over land. Figure 6 shows that the inclusion ofthe effects of land improves the Atlantic forecasts byabout 15% at 24–48 h. These improvements were sta-tistically significant out to 72 h. In contrast, the eastPacific forecast improvements were 3% or less, andnone was statistically significant. This result is not sur-prising since the east Pacific storms interact with landmuch less frequently than do Atlantic storms. The in-

clusion of land does not significantly improve the At-lantic forecasts at day 4 and results in small (but notstatistically significant) degradation at 120 h. It is likelythat the track forecast errors at these longer rangesintroduce a source of error that cancels any improve-ment due to the inclusion of land effects.

The SHIPS forecasts were extended to 5 days begin-ning in 2001, using the same methodology as for 12–72h. Figure 7 shows the mean absolute intensity errors for2001–03 from D-SHIPS and SHIFOR5. For the Atlan-tic, the D-SHIPS intensity errors grow approximatelylinearly with time, but the SHIFOR5 errors level offafter 72 h. By 5 days the Atlantic D-SHIPS errors arelarger than SHIFOR5, indicating no skill. A possible

FIG. 5. The skill (relative error) for homogeneous samples ofthe SHIPS and NCEP/GFDL intensity forecasts for the Atlanticand eastern North Pacific for 1997–2003.

FIG. 6. The percent improvement (reduction in mean forecasterror) of the decay version of SHIPS relative to the version with-out the decay factor. All cases from 2000 to 2003 are includedusing the “new” NHC verification rules.

FIG. 7. The average forecast errors (kt) of the 5-day versions ofthe SHIFOR model and decay SHIPS model for a homogeneoussample from 2001 to 2003 for the Atlantic and eastern NorthPacific.


explanation for the continued growth of the D-SHIPSerrors is that the track errors make a significant contri-bution to the intensity prediction. The median error ofthe 5-day Atlantic track forecast used by SHIPS wasabout 520 km, which is a significant fraction of the outerradius of the area over which most of the predictors arecalculated. To further investigate the issue, the 5-daytrack errors were correlated with the 5-day intensityerrors. This analysis confirmed that there is a positivecorrelation between the track and intensity errors.

In contrast to the Atlantic, the east Pacific D-SHIPSerrors in Fig. 7 level off after 72 h in a manner similarto that of the SHIFOR5 errors, and are less than theSHIFOR5 errors at days 4 and 5, indicating forecastskill. The median 5-day track error for the east Pacificwas 450 km, which is less than that of the Atlantic track.Thus, the impact of track errors on intensity errors isless in the east Pacific. Also, the most severe impact ofthe track error on the intensity forecast occurs whenlandfall is forecast but does not occur or vise versa.There are considerably fewer storms that move nearland in the east Pacific, which may also explain why themodel has skill at the longer time intervals in the eastPacific.

From a user perspective the very large intensity er-rors are the most problematic. To give a better idea ofthe tails of the distributions, the 95th percentile of theD-SHIPS, SHIFOR5, and official NHC errors werecomputed for the sample with 5-day forecasts (2001–03)using the new verification rules (Table 2). Also shownis the RE of the mean and 95th percentile of the D-SHIPS and official forecasts. This table shows that D-SHIPS improves upon the climatology and persistencemean error by up to 22.5% at 12–72 h, again indicating

forecast skill. The official forecasts were skillful out to5 days. The RE of the 95th percentile error is similar tothat of the mean error. Thus, D-SHIPS improves themost difficult forecasts by about the same percentage asthe total forecast sample. Table 2 also shows that the95th percentile D-SHIPS and official errors are quitelarge. This result again indicates that considerable im-provement in intensity forecasting is still needed.

The 95th percentiles of the east Pacific forecasts werealso examined for the 2001–03 sample using the newverification rules (not shown). Results were fairly simi-lar to those from the Atlantic in that the percent im-provements of the 95th percentile error were similar tothe mean error improvements.

4. Inclusion of satellite data

Most of the predictors in Table 1 are measures of thestorm environment so there is little direct informationabout the storm itself. To provide additional storm in-formation, an experimental version of SHIPS was de-veloped that includes predictors from GOES infraredimagery. The GOES data should provide informationabout the structure of the deep convection near thestorm center. Cloud-top brightness temperatures donot always directly relate to convective updrafts on a“pixel by pixel” basis due to sloping eyewalls, cirrusoutflow, and other factors. However, validation of in-frared-based precipitation methods indicates that thereis a statistical relationship between the area-integratedcold cloud tops and precipitation in the storm innercore (e.g., Scofield et al. 2001). Also, the lack of coldcloud-top temperatures is well correlated with the lackof deep convection.

TABLE 2. The mean absolute error (kt) and the 95th percentile of the absolute error (kt) of the Atlantic SHIFOR5, D-SHIPS, andofficial forecasts for the 2001–03 sample. Also shown are the relative errors (%) of the D-SHIPS and official mean and 95th percentileerrors.

Forecast interval (h)

12 24 36 48 72 96 120

SHIFOR5Mean error 7.7 12.0 15.6 18.4 22.0 24.1 23.095th percentile error 22 33 41 45 50 55 52

D-SHIPSMean error 6.7 9.3 12.1 14.8 19.5 24.0 27.095th percentile error 18 25 31 36 46 58 62Mean RE 13.6 22.3 22.5 19.5 11.1 0.4 �17.795th percentile RE 18.2 24.2 24.4 20.0 8.0 �5.5 �19.2

OfficialMean error 5.6 9.0 11.4 14.2 18.2 19.9 21.295th percentile error 15 25 30 35 45 55 60Mean RE 26.8 25.1 27.3 22.8 17.3 17.6 8.095th percentile RE 31.8 24.2 26.8 22.2 10.0 0.0 �15.3

Sample size 947 852 765 683 536 428 347


The GOES data were obtained from the archivemaintained by the Cooperative Institute for Researchin the Atmosphere (CIRA). Since 1995, channel 4 in-frared (10.7 �m) imagery from GOES-East and GOES-West for nearly all Atlantic and east Pacific tropicalcyclones has been collected (Zehr 2000). The channel 4brightness temperatures (TB) were azimuthally aver-aged on a 4-km, storm-centered radial grid. The TB

standard deviations from the azimuthal average werealso calculated at each radius. In all cases, the GOESdata were within 1 h of the time from the correspondingSHIPS case.

Several studies have shown that the oceanic heat con-tent (OHC) relative to the depth of the 26°C isothermis important for tropical cyclone intensity change (e.g.,Shay et al. 2000). Through the retrieval and analysis ofhigh-resolution blended surface height anomaly (SHA)altimetry data, the Reynolds SST weekly analysis, andin combination with a hurricane season (June–November) averaged oceanic climatology, the OHCrelative to 26°C water was estimated by applying theapproach described by Mainelli-Huber (2000). OHCanalyses in the prestorm environments of most Atlanticcases were available back to 1995 over a domain thatextends from 0° to 40°N and 100° to 50°W and over theentire Atlantic basin since 2002. OHC analyses are notyet available for the eastern Pacific.

The incorporation of SHA altimetry data into OHCestimations has changed over the last decade. Prior to1999 only TOPEX/Poseidon (T/P) altimeter data wereoperationally accessible. The T/P satellite provides a10-day repeat cycle; however, data are spaced up to 3°apart in longitude. Earlier studies by Mainelli-Huber(2000) revealed that the poor spatial resolutions would,at times, cause a misrepresentation of oceanic features.During the summer of 2002 the T/P was replaced byJason-1, which provides the same temporal and spatialresolution.

The European Remote Sensing Satellite-2 (ERS-2) ra-dar altimetry data became available in 1999. The ERS-2provided a higher spatial resolution dataset with a 35-day repeat cycle. In late June of 2003, the ERS-2 be-came inoperable. However, most of the experimentalSHIPS runs occurred while both T/P (or Jason-1) andERS-2 were available. The 10-day T/P (or Jason-1) and15-day ERS-2 data were objectively analyzed followingthe approach of Mariano and Brown (1992), which pro-vides estimates on an evenly spaced 0.5° latitude–longitude grid. Currently, the Geosat Follow-On(GFO) altimeter data (with a 17-day repeat cycle) havebecome fully operational and have replaced the ERS-2.All OHC estimates incorporated into SHIPS are now ablend of Jason-1 and GFO. The real-time OHC analy-

ses are updated daily at NHC using the previous 10days of altimetry data over a domain that includes theentire Atlantic basin to 60°N.

If the satellite predictors were combined with theother SHIPS predictors, it would be necessary to re-duce the sample size because these new data were onlyavailable back to 1995, and over only a limited part ofthe Atlantic basin. To overcome this problem, a two-step prediction procedure is applied. In the first step theSHIPS model with the full data sample is derived asbefore. Then, the difference between the SHIPS pre-dictions and the observed intensity changes are used asthe dependent variables in a second multiple regres-sion, where parameters from the GOES and altimetrydata are the independent variables. This second regres-sion will be referred to as the perturbation model.

The perturbation model includes two GOES predic-tors and one OHC predictor. The GOES predictors are1) the percent of the area (pixel count) from 50 to 200km from the storm center where TB is colder than�20°C and 2) the standard deviation of TB (relative tothe azimuthal average) averaged from 100 to 300 km.Several other GOES predictors were tested, but thetwo described above explained the most variance in theperturbation model. The GOES standard deviationpredictor is actually the product of the TB standarddeviation and the initial intensity. This variable pro-vided a better fit in the perturbation model and helpsaccount for the variability of the TB standard deviationas a function of storm intensity. The altimetry predictoris the OHC above 50 KJ cm�2, averaged along thestorm track. A better fit of the perturbation model wasobtained using this threshold. In fact, the OHC predic-tor was not statistically significant without the thresh-old. The GOES parameters are static predictors andthe altimetry parameter is a time- dependent predictor.The OHC generally exceeds 50 KJ cm�2 in the Carib-bean, the Loop Current, in warm-core eddies in theGulf of Mexico, and along the Gulf Stream. For theeastern Pacific model, only the two GOES predictorswere included in the perturbation model.

The perturbation model was run in real time duringthe 2003 hurricane season for all Atlantic and east Pa-cific forecast cases, but was not ready in time for the2002 season. In the 2002–03 evaluations described be-low, the 2003 cases are real-time forecasts and the 2002cases are postseason runs that used input that would beavailable in real time (GFS forecasts and NHC opera-tional tracks and initial intensity estimates).

Figure 8 shows the percent improvement in the D-SHIPS forecasts for the 2002–03 samples due to thesatellite data. For the east Pacific, the inclusion of theGOES predictors improves the forecasts by as much as


7.4% at 12–72 h. All of these improvements were sta-tistically significant. By 120 h, the east Pacific forecastswith the GOES data were slightly degraded, but thedifferences in the average errors were not statisticallysignificant. In contrast to the east Pacific, the inclusionof the satellite data (GOES and OHC predictors) in theAtlantic resulted in only a very slight improvement at12–48 h and some degradation at the longer forecastintervals. None of the Atlantic improvements was sta-tistically significant.

Because the GOES data improved the east Pacificforecasts but the combination of OHC and GOES datahad a neutral effect on the Atlantic forecasts, it mightbe suspected that the OHC predictor was canceling thepositive impact of the GOES data on the Atlantic fore-casts. To investigate this possibility, all of the Atlanticforecasts were rerun with the contributions to the per-turbation intensity changes from the OHC predictor setto zero. This procedure was repeated but with the con-tributions from the GOES predictors set to zero. Be-cause of potential cancellation between the impacts ofthe OHC and GOES predictors, the sum of the im-provements from the forecasts with only GOES andonly OHC predictors does not add up to the improve-ments when both are included. However, this analysisillustrates whether one or the other of the predictors isdominating the improvements or degradations. Resultsshowed that the improvements in the short-range At-lantic forecasts (12–24 h) and degradations at thelonger periods were roughly evenly divided betweenthe contributions from the GOES and OHC predictors.Thus, it does not appear that the OHC predictors wereinterfering with the GOES predictors.

The perturbation SHIPS model for the Atlantic wasdeveloped from cases that were nearly all west of 50°W

due to the limitation in the size of the OHC analysisarea. However, the perturbation model was applied toall Atlantic cases in the operational forecasts from 2002to 2003. To determine if this mismatch between thedevelopmental and operational samples had any im-pact, the 2002–03 forecasts were verified for only thosecases where the storm was west of 50°W. The improve-ments for this subsample in Fig. 8 are nearly twice aslarge as for the total sample at 12–48 h, although theforecasts were still degraded at 96 and 120 h. Similar tothe total sample, the GOES and OHC data contributedabout equally to the short-range improvements andlonger-range degradations. This result suggests that aneffort should be made to obtain OHC analyses over theentire Atlantic basin or restrict the application of theperturbation model to the same area as used to developthe model. Efforts are under way to obtain the clima-tology of OHC analyses over the entire Atlantic basinfor the developmental sample.

Figure 8 shows that the satellite data degraded theAtlantic and east Pacific forecasts at 4 and 5 days. Theversion of SHIPS with the satellite data included be-came operational for the 2004 season. Results from apreliminary verification for the Atlantic and east Pacific(through the end of October 2004) are very encourag-ing and are even better than what was shown in Fig. 8.The improvements with the satellite data are similar tothose in Fig. 8 out to 72 h, but there is no negativeimpact at 96 and 120 h. The developmental datasets forthe 2004 perturbation model had much larger samplesizes at 4 and 5 days due to the addition of the 2002 and2003 cases, which likely helped to eliminate the degra-dation of the longer-range forecasts.

5. Concluding remarks

Modifications to the NHC operational SHIPS inten-sity model for each year from 1997 to 2003 were de-scribed. Major changes include the addition of amethod to account for the storm decay over land in2000, the extension of the forecasts from 3 to 5 days in2001, and the replacement of a simple dry-adiabaticmodel with the NCEP operational global model for theevaluation of the atmospheric predictors, beginning in2001. Several new predictors were also described. Veri-fication of the SHIPS forecasts showed the followingresults: 1) for the period 1997–2003, the SHIPS fore-casts had statistically significant skill (relative to fore-casts based upon climatology and persistence) out to 72h in the Atlantic, and at 48 and 72 h in the east Pacific;2) the operational NCEP dynamical intensity forecast(GFDL) model did not have any significant forecastskill for this same period; 3) The 4- and 5-day SHIPS

FIG. 8. The percent improvement of the SHIPS intensity fore-casts with satellite input relative to the operational version with-out the satellite data for the 2002–03 samples in the east Pacific,the Atlantic, and the Atlantic for storms initially west of 50°W.


forecasts that began in 2001 were not skillful in theAtlantic but showed some modest skill in the east Pa-cific; 4) the inclusion of the effects of the decay overland beginning in 2000 reduced short-range Atlantic(East Pacific) intensity errors by up to 15% (3%) buthad a neural impact after 72 h; 5) the mean and 95thpercentiles of the SHIPS intensity errors were docu-mented.

An experimental version of SHIPS that included sat-ellite observations was tested during the 2002 and 2003seasons. New predictors included brightness tempera-ture information from GOES channel 4 (10.7 �m) im-agery, and oceanic heat content (OHC) estimates in-ferred from satellite altimetry observations. The OHCestimates were only available for the Atlantic basin.The GOES data significantly improved the east Pacificforecasts by up to 7% at 12–72 h. The combination ofGOES and satellite altimetry improved the Atlanticforecasts by up to 3.5% through 72 h for those stormswest of 50°W.

To provide perspective on the current state of inten-sity forecasting, the skill of the NHC official intensityforecasts was compared with the track forecast skill.The Atlantic intensity and track skill are comparable at12 h, but by 36 h (72 h) the track skill is twice (threetimes) the intensity forecast skill. Thus, intensity fore-casting still has a long way to go.

Several modifications are planned for the SHIPSmodel to improve its performance. Methods are beingdeveloped to include predictors from aircraft flight-level observations. It is anticipated that the inner-corewind field structure information will provide additionalshort-range forecast skill for those storms that have po-tential to make landfall along the U.S. coastline. TheGOES data are currently included in a very simple way(pixel counts and brightness temperature standard de-viations averaged over large areas). The radial patternsof the GOES data near the storm center are being ex-amined in greater detail using an empirical orthogonalfunction (EOF) approach. A daily SST analysis with a50-km horizontal resolution under development atNCEP is being compared with the weekly 100-km Rey-nold’s SST analyses currently used by SHIPS (Berg etal. 2004). In addition, experimental daily SST analysesfrom the Advanced Microwave Scanning Radiometerfor the Earth Observing System (AMSR-E) instrumentare showing potential for improving the SHIPS fore-casts, especially in the east Pacific basin. Another po-tential use of satellite observations is to improve theanalysis of the storm thermodynamic environment.Dunion and Velden (2004) have shown that total pre-cipitable water analysis from microwave imagery hasthe potential to provide a better method for identifying

the SAL than current global model analyses. In thelonger term, new sounders planned for the next-generation National Oceanic and Atmospheric Admin-istration (NOAA) polar-orbiting and geostationary sat-ellite systems should also be of utility. Finally, neuralnetwork techniques are being explored to determine ifthey can improve the SHIPS forecasts, as suggested byBaik and Paek (2000) for the western North Pacific.

Acknowledgments. This work was partially supportedby a grant to CIRA from the NOAA U.S. WeatherResearch Program Joint Hurricane Testbed. L. K. Shaywas supported by the National Science Foundationthrough Grant ATM-01-08218. Tom Cook helped todevelop the real-time satellite-derived heat contentanalyses and Inger Solheim and James Kossin assistedwith the processing of the GOES data. Jack Dostalek,Miles Lawerence, Richard Pasch, Fran Holt, and threeanonymous reviewers provided useful comments on anearlier version of the manuscript. The views, opinions,and findings in this report are those of the authors andshould not be construed as an official NOAA and/orU.S. government position, policy, or decision.

REFERENCES

Baik, J.-J., and J.-S. Paek, 2000: A neural network model forpredicting typhoon intensity. J. Meteor. Soc. Japan, 78, 857–869.

Berg, R., C. Sisko, and M. DeMaria, 2004: High resolution SST inthe SHIPS model: Improving operational guidance of tropi-cal cyclone intensity forecasts. Preprints, 26th Conf. on Hur-ricanes and Tropical Meteorology, Miami, FL, Amer. Meteor.Soc., 536–537.

DeMaria, M., and J. Kaplan, 1994a: A statistical hurricane inten-sity prediction scheme (SHIPS) for the Atlantic basin. Wea.Forecasting, 9, 209–220.

——, and ——, 1994b: Sea surface temperature and the maximumintensity of Atlantic tropical cyclones. J. Climate, 7, 1324–1334.

——, and ——, 1999: An updated statistical hurricane intensityprediction scheme (SHIPS) for the Atlantic and easternNorth Pacific basins. Wea. Forecasting, 14, 326–337.

——, and J. M. Gross, 2003: Evolution of tropical cyclone forecastmodels. Hurricane! Coping with Disaster, R. Simpson, Ed.,Amer. Geophys. Union, 103–126.

Dunion, J. P., and C. S. Velden, 2004: The impact of the Saharanair layer (SAL) on Atlantic tropical cyclone activity. Bull.Amer. Meteor. Soc., 85, 353–364.

Fitzpatrick, P. J., 1997: Understanding and forecasting tropical cy-clone intensity change with the Typhoon Intensity PredictionScheme (TIPS). Wea. Forecasting, 12, 826–846.

Horsfall, F. M., M. DeMaria, and J. M. Gross, 1997: Optimal useof large-scale boundary and initial fields for limited-area hur-ricane forecast models. Preprints, 22d Conf. on Hurricanesand Tropical Meteorology, Fort Collins, CO, Amer. Meteor.Soc., 571–572.

Jarvinen, B. R., and C. J. Neumann, 1979: Statistical forecasts of


tropical cyclone intensity change. NOAA Tech. Memo. NWSNHC-10, 22 pp.

Kalnay, E., 2003: Atmospheric Modeling, Data Assimilation andPredictability. Cambridge University Press, 341 pp.

Kaplan, J., and M. DeMaria, 1995: A simple empirical model forpredicting the decay of tropical cyclone winds after landfall.J. Appl. Meteor., 34, 2499–2512.

——, and ——, 2001: On the decay of tropical cyclone winds afterlandfall in the New England area. J. Appl. Meteor., 40, 280–286.

Knaff, J. A., M. DeMaria, C. R. Sampson, and J. M. Gross, 2003:Statistical 5-day tropical cyclone intensity forecasts derivedfrom climatology and persistence. Wea. Forecasting, 18, 80–92.

Kurihara, Y., R. E. Tuleya, and M. A. Bender, 1998: The GFDLHurricane Prediction System and its performance in the 1995hurricane season. Mon. Wea. Rev., 126, 1306–1322.

Mainelli-Huber, M., 2000: On the role of the upper ocean in tropi-cal cyclone intensity change. M.S. thesis, Dept. of Meteorol-ogy and Physical Oceanography, University of Miami, 73 pp.

Mariano, A. J., and O. B. Brown, 1992: Efficient objective analysisof heterogeneous and nonstationary fields via parameter ma-trix. Deep-Sea Res., 7, 1255–1271.

McAdie, C. J., and M. B. Lawrence, 2000: Improvements in tropi-

cal cyclone track forecasting in the Atlantic basin, 1970–98.Bull. Amer. Meteor. Soc., 81, 989–998.

Neumann, C. J., 1972: An alternate to the HURRAN tropicalcyclone forecast system. NOAA Tech. Memo. NWS SR-62,22 pp.

Reynolds, R. W., and T. M. Smith, 1993: An improved real-timeglobal sea surface temperature analysis. J. Climate, 6, 114–119.

Scofield, R. A., M. DeMaria, and R.M del Alfaro, 2001: Space-based rainfall capabilities in hurricanes offshore and inland.Preprints, Symp. on Precipitation Extremes: Prediction, Im-pacts and Responses, Albuquerque, NM, Amer. Meteor. Soc.,297–301.

Shay, L. K., G. J. Goni, and P. G. Black, 2000: Effects of a warmoceanic feature on Hurricane Opal. Mon. Wea. Rev., 128,1366–1383.

Whitney, L. D., and J. S. Hobgood, 1997: The relationship be-tween sea surface temperatures and maximum intensities oftropical cyclones in the eastern North Pacific Ocean. J. Cli-mate, 10, 2921–2930.

Zehr, R. M., 2000: Tropical cyclone research using large infrareddata sets. Preprints, 24th Conf. on Hurricanes and TropicalMeteorology, Fort Lauderdale, FL, Amer. Meteor. Soc., 486–487.


Documents

Further Improvements to the Statistical Hurricane ...rammb.cira.colostate.edu/resources/docs/further_demaria.pdfA verification of the SHIPS operational intensity forecasts is presented